If $T$ is self-adjoint on finite dimensional vector space $V$, then $\text{det} (T-tI_{V})$ splits over $\mathbb{R}$.

I would like to know "why"?

Notice: here we use this Lemma for proving the diagonalization theorem or another form of spectral theorem for self-adjoint operators, so we need a direct proof as @RobertLewis provided.


Suppose $T$ is linear transformation on $V$. $T$ is a self-adjoint operator if and only if there exists an orthonormal basis such that $[T]_{\beta}$ is diagonal.


($\Rightarrow$) $T$ is self-adjoint. Then, $\text{det} (T-tI_{V})$ splits over $\mathbb{R}$. Then, by Schur theorem there exists an orthonormal basis $\beta$ such that $[T]_{\beta}$ is uppertriangular. We have: $[T]_\beta = ([T]_{\beta})^{*}$. Hence, $[T]_{\beta}$ is diagonal.


It is clear that we use the above Lemma to prove this theorem not conversely!

  • 1
    $\begingroup$ Spectral theorem $\endgroup$ – user223391 Nov 7 '17 at 6:24
  • $\begingroup$ @ZacharySelk: we use this Lemma to prove the spectral theorem about self-adjoint operators NOT conversly. $\endgroup$ – Amin Nov 8 '17 at 1:41

The eigenvalues of a self-adjoint operator such as $T$ are all real: if

$T v = \lambda v, \tag 1$


$\lambda \langle v, v \rangle = \langle v, \lambda v \rangle = \langle v, Tv \rangle, \tag 2$


$\bar \lambda \langle v, v \rangle = \overline{\langle v, Tv \rangle} = \langle Tv, v \rangle = \langle v, T^\dagger v \rangle = \langle v, Tv \rangle = \lambda \langle v, v \rangle, \tag 3$

whence, since $\langle v, v \rangle \ne 0$,

$\bar \lambda = \lambda \in \Bbb R. \tag 4$

Since the eigenvalues $\lambda_i \in \Bbb R$ of $T$ are the roots of the $\deg(\dim V)$ polynomial

$p_T(t) = \det(T - t I_V), \tag 5$

we have

$p_T(t) = \det(T - tI_V) = \displaystyle \prod_1^{\dim V}(\lambda_i - t), \tag 6$

i.e, $p_T(t)$ is the product of $\dim V$ linear factors $\lambda_i - t \in \Bbb R[t]$; thus $p_T(t)$ splits in $\Bbb R[t]$, that is, over $\Bbb R$.

The "why" is essentially that $T = T^\dagger$ forces all $\dim V$ of the $\lambda_i \in \Bbb R$, leading to one real factor $\lambda_i - t$ for each of the $\dim V$ eigenvalues.

Note: The assumption that $\dim V < \infty$ is used to affirm the existence of $p_T(t) = \det(t - t I_V)$. End of Note.

  • 1
    $\begingroup$ Thanks @RobertLewis. I did not pay attention to the equation (6)! $\endgroup$ – Amin Nov 7 '17 at 6:59
  • 1
    $\begingroup$ My pleasure, sir! $\endgroup$ – Robert Lewis Nov 7 '17 at 7:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.